Version 2.5.2 of OpenMx now available

The newest release of OpenMx, version 2.5.2, is now available through CRAN and through our own repository.

New features include:

• Regression factor-score estimates are now available for RAM path models via mxFactorScores().
• mxGenerateData() can now generate data conditional on definition variables.
• SLSQP is now capable of using an analytic gradient during optimization.
• Numerous substantial improvements have been made to mxTryHard(). In particular, there are now four additional wrapper functions--mxTryHardOrig(), mxTryHardSSCT(), mxTryHardWideSearch(), and mxTryHardOrdinal()--which have default values for certain arguments that are tailored toward a specific purpose.
• A new function, imxRobustSE(), which calculates robust standard errors for parameter estimates, from the "sandwich estimator."
• Some functions have been newly made usable in MxAlgebras: the inverse trigonometric functions, the inverse hyperbolic functions, logp2z() (standard-normal quantile function from log probabilities), lgamma1p() (accurate lgamma(x+1) for small x), the Bessel functions, dbeta(), and pbeta(). The latter two are prototypes for making the 'd' and 'p' probability-distribution functions from the 'stats' package usable in MxAlgebras.

Bug fixes and performance tweaks include:

• Two GREML-related bugs have been repaired. One pertained to the behavior of mxGREMLDataHandler() when blockByPheno=FALSE. The other pertained to the mxGREML feature's automated handling of missing data when some of the derivatives of the 'V' matrix are MxAlgebras.
• LISREL path models now handle means correctly.
• mxFactorScores() now returns factor scores in row ordering of the original raw data, and not in the row ordering of the auto-sorted data.
• The known issue from the release announcement of v2.3.1, which involved factor-score estimates on factors with nonzero means, has been resolved.
• Using mxFactorScores() with type="ML" or type="WeightedML" no longer fails with an error when standard errors are not available from the input model.
• Several help pages have been updated, clarified, and made more complete.
• Changes were made to OpenMx's internal interface with NPSOL to ensure that optimizer consistently respects the value of the "Major iterations" option.
• The behavior of the Newton-Raphson optimizer when it encounters a parameter bound has been improved, and should result in fewer convergence failures.
• mxGenerateData() now works properly with continuous-time state-space models.
• The sufficient statistic likelihood was adjusted to match the full information likelihood value. Prior versions of OpenMx (and all the way back to Mx), used a slightly different formula that did not correspond exactly to the full information formula.